GENETIC ALGORITHM BASED SENTENCE EXTRACTION FOR TEXT

SUMMARIZATION

Ladda Suanmali1, Naomie Salim2, Mohammed Salem Binwahlan3

1Faculty of Science and Technology, Suan Dusit Rajabhat University

Dusit, Bangkok, 10300 Thailand

2Faculty of Computer Science and Information Systems,

Universiti Teknologi Malaysia, 81310 Skudai, Johor, Malaysia

3Faculty of Applied Sciences, Hadhramout University of Science and Technology

Seiyun, Hadhramout, Yemen

email: 1nongnu_@hotmail.com, 2naomie@utm.my, 3moham2007med@yahoo.com

ABSTRACT

The goal of text summarization is to generate summary of the original text that helps

the user to quickly understand large volumes of information available in that text. This

paper focuses on text summarization based on sentence extraction. One of the methods

to obtain suitable sentences is to assign some numerical measure for sentences called

sentence weighting and then select the best ones. The first step in summarization by

extraction is the identification of important features. In this paper, we consider the

effectiveness of the features selected using Genetic Algorithm (GA). GA is used for the

training of 100 documents in DUC 2002 data set to learn the weight of each feature,

which is evaluated using recall measurement generated by ROUGE for a fitness

function. The weights obtained by GA were used to adjust the important features score.

We compare our results with Microsoft Word 2007 summarizer and Copernic

summarizer both for 100 documents and 62 unseen documents. The results show that

the best average precision, recall, and f-measure for the summaries were obtained by

GA.

Key Word: Genetic Algorithm, Sentence extraction, Statistic method, Text

summarization

.

1.0 INTRODUCTION

Text summarization has become an important and timely tool for interpreting large

volumes of text available in documents. One of the natural questions to ask in doing

summarization is “what are the properties of text that should be represented or kept in a

summary?”

Text summarization addresses both the problem of finding the most important subset of

sentences in text, which in some way represents its source text and the problem of

generating coherent summaries. This process is significantly different from human

based text summarization since human can capture and relate deep meanings and

themes of text documents while automation of such a skill is very difficult to

implement. The goal of text summarization is to present the most important information

in a shorter version of the original text while keeping its main content and helps the user

to quickly understand the large volume of information. Automatic text summarization

researchers since Luhn (1958) are trying to solve or at least relieve that problem by

proposing techniques for generating summaries. A number of researchers have

proposed techniques for automatic text summarization which can be classified into two

categories: extraction and abstraction. Extraction summary is produced by selecting

sentences or phrases from the original text with the highest score and put it together into

a new shorter text without changing the source text. Abstraction summary method uses

linguistic methods to examine and interpret the text for generative of abstracts. Most of

the current automated text summarization systems use extraction method to produce a

summary (Ko and Seo, 2008; Yulia et al., 2008; Suanmali et al., 2009; Ramiz, 2009).

Sentence extraction techniques are commonly used to produce extraction summaries.

One of the methods to obtain suitable sentences is to assign some numerical measure of

a sentence for the summary called sentence weighting and then select the best sentences

to form document summary based on the compression rate. In the extraction method,

compression rate is an important factor used to define the ratio between the length of

the summary and the source text. As the compression rate increases, the summary will

be larger, and more insignificant content is contained. While the compression rate

decreases the summary to be short, more information is lost. In fact, when the

compression rate is 5-30%, the quality of summary is acceptable (Fattah and Ren, 2008;

Yeh et al., 2005; Mani and Maybury, 1999; Kupiec et al., 1995).

The first step in summarization by extraction is the identification of important features

such as sentence length, sentence location (Fattah and Ren, 2008), term frequency

(Salton, 1989), number of words occurring in title (Salton and Buckley, 1997), number

of proper nouns (Yulia et al., 2008) and number of numerical data (Lin, 1999). In our

approach, we utilize a feature fusion technique to discover which features out of the

available ones are most useful.

Kiani and Akbarzadeh (2006) proposed technique for summarizing text using a

combination of Genetic Algorithm (GA) and Genetic Programming (GP) to optimize

rule sets and membership function of fuzzy systems. Vahed et al. (2008) used GA to

produce document summary. They proposed a fitness function based on three following

factors: Readability Factor (RF), Cohesion Factor (CF) and Topic-Relation Factor

(TRF).

In this paper, we propose use genetic algorithm method to extract important sentences

as a summary. The rest of this paper is organized as follows. Section 2 describes

preprocessing and the important features used. Section 3 and 4 describes our proposed

method, followed by experimental design, experimental results and evaluation. Finally,

we conclude and suggest future work that can be carried out in Section 5.

2.0 EXPERIMENTAL DESIGN

We used the test document sets (D061j, D062j, D063j, D064j, D065j, D066j, D067f,

D068f, D069f, D070f, D071f, D072f, D073b, D074b, D075b, D077b, D078b, D079b,

and D080b) from DUC2002 (DUC., 2002) comprising of 162 documents to create

automatic document summarization. Each document in DUC2002 collection is supplied

with a set of human-generated summaries provided by two different experts. While

each expert was asked to generate summaries of different length, we use only generic

100-word variants.

We divide the 162 documents into two groups. The first 100 documents were used for

training. The other 62 documents were used to evaluate and compare the results.

Currently, input document is of plain text format. There are four main activities

performed in this stage: sentence segmentation, tokenization, stop word removal, and

word stemming. Sentence segmentation is performed by boundary detection and

separating source text into sentences. Tokenization is done by separating the input

document into individual words. Next, words which rarely contribute to useful

information in terms of document relevance and appear frequently in document but

provide less meaning in identifying the important content of the document are removed.

These words include articles, prepositions, conjunctions and other high-frequency

words such as ‘a’, ‘an’, ‘the’, ‘in’, ‘and’, ‘I’, etc... The last step for preprocessing is

word stemming. Word stemming is the process of reducing inflected or derived words

to their stem, base or root form. In this research, we performed words stemming using

Porter’s stemming algorithm (Porter, 1980). For example, a stemming algorithm for

English should stem from the words ‘compute’, ‘computed’, ‘computer’, ‘computable’,

and ‘computation’ to its word stem, ‘comput’.

2.1 SENTENCE FEATURES

After preprocessing, each sentence of the document is represented by an attribute vector

of features. These features are attributes that attempt to represent the sentence in the

task of sentence selection. We focus on eight features for each sentence. Each feature is

given a value between ‘0’ and ‘1’. There are eight features as follows:

2.1.1 Title feature

The word in sentence that also occurs in the title is given higher score. This is

determined by counting the number of matches between the content words in a sentence

and the words in the title. We calculate the score for this feature which is the ratio of the

number of words in the sentence that occur in the title over the number of words in title.

2.1.2 Sentence Length

This feature is useful to filter out short sentences such as datelines and author names

commonly found in news articles. The short sentences are not expected to belong to the

summary. We use normalized length of the sentence, which is the ratio of the number of

words occurring in the sentence over the number of words occurring in the longest

sentence of the document.

S_F2(S) = No.Word occurring in S (2)

No.Word occurring in longest sentence

S_F1(S) = No.Title word in S (1)

No.Word in Title

2.1.3 Term Weight

The frequency of term occurrences within a document has often been used for

calculating the importance of sentence. The score of a sentence can be calculated as the

sum of the score of words in the sentence. The score wi of word i can be calculated by

the classic tf.idf method as follows (Salton and Buckley, 1997). We applied this method

to tf.isf (Term frequency, Inverse sentence frequency).

where tfi is the tern frequency of word i in the document, N is the total number of

sentences, and ni is number of sentences in which word i occurs. This feature can be

calculated as follows.

where k is number of words in sentence.

2.1.4 Sentence Position

A sentence position in the text can indicate importance of the sentence. This feature can

involve several items such as the position of a sentence in the document, section, and

paragraph, etc.. The first sentence has the highest ranking. We only consider up to 5

positions from the top of the document. For instance, the first sentence in a paragraph

has a score value of 5/5, the second sentence has a score 4/5, and so on.

2.1.5 Sentence to Sentence Similarity

Similarity between sentences, for each sentence s, is the similarity between s and all

other sentences, as computed by the cosine similarity measure. The score of this feature

for a sentence s is obtained by computing the ratio of the summary of sentence

similarity of sentence s with all other sentences over the maximum of sentence

similarity.

2.1.6 Proper Noun

Usually the sentence that contains more proper nouns is an important one and it is most

probably included in the document summary. The score for this feature is calculated as

the ratio of the number of proper nouns in the sentence over the sentence length.

2.1.7 Thematic Word

The number of thematic word in a sentence is important because terms that occur

frequently in a document are probably related to the same topic. The number of

thematic words indicates the words with maximum possible relativity. We used the top

S_F6(S) = No. Proper nouns in S (7)

Length (S)

S_F5(S) = Sum of Sentemce Similarity in S (6)

Max(Sum of Sentence Similarity)

S_F4(S) = 5/5 for 1st, 4/5 for 2nd, 3/5 for 3rd,

2/5 for 4th, 1/5 for 5th,

0/5 for other sentences (5)

10 most frequent content word for consideration as thematic. The score for this feature

is calculated as the ratio of the number of thematic words in the sentence over the

maximum summary of thematic words in the sentence.

2.1.8 Numerical Data

A sentence that contains numerical data is considered important and is most probably

included in the document summary. The score for this feature is calculated as the ratio

of the number of numerical data in sentence over the sentence length.

2.2 THE METHODS

The features score of each sentence can be calculated as described in the previous

section. In this section, we use two methods to extract important sentences: text

summarization based on general statistic method (GSM) and Genetic Algorithm (GA).

2.2.1 Text Summarization based on General Statistic Method (GSM)

The feature score of each sentence described in the previous section are used to obtain

the significant sentences. In this section, we used general statistic method (GSM) to

extract the important sentences. The technique consists of the following main steps.

(1) Read the source document into the system.

S_F8(S) = No. Numerical data in S (9)

Length (S)

S_F7(S) = No. Thematic word in S (8)

Max(No. Thematic word)

(2) For preprocessing, the system extracts the individual sentences of the original

documents. Then, the input document is separated into individual words. Next, remove

the stop words. The last step of preprocessing is word stemming.

(3) Each sentence is associated with a vector of eight features described in Section 2.1,

whose values are derived from the content of the sentence.

(4) The features are calculated to obtain the sentence score based on general statistic

method (GSM) shows in Figure 1;

(5) A set of the highest score sentences are extracted as a document summary based on

the compression rate.

Figure 1: Text summarization based on general statistic method architecture

Text summarization based on general statistical method is produced by using the

sentence weight. First, for a sentence s, a weighted score function, as shown in the

following equation (eq. 10) is exploited to integrate all the eight feature scores

mentioned in Section 2.1.

Preprocessing

Source

Document

Extraction of

Features

Calculation of

Sentence Score

Extraction of

Sentences

Summary

Document

Where Score(S) is the score of the sentence S and S_Fk(S) is the score of the feature k

2.2.1 Text Summarization based on Genetic Algorithm (GA)

Genetic algorithm (GA) provides an alternative to traditional optimization techniques

by using directed random searches to locate optimal solutions in complex landscapes.

GA generates a sequence of populations by using a selection mechanism, where cross-

overs and mutations are used as part of the search mechanisms (Srinivas and Patnaik,

1994), as shown in Figure 2.

Figure 2: Simple genetic algorithm structure

Initial a population

of chromosomes

Fitness Evaluation

Select next generation

Perform reproduction using

crossover and mutation

End of Criteria? No Yes

Results obtained

Chromosome Encoding

Before applying a GA, we first must encode the parameters of the problem to be

optimized. GAs work with codes that represent the parameter. There are two common

representation methods; floating point and bit string that can be used to represent the

parameter. In our research, we use the bit string (containing binary digits: 0s and 1s)

because the majority of genetic operators are suitable for our representation. To learn

the feature weights, each chromosome contains the genes connected together into a long

string represented by a binary vector of dimension F (where F is the total number of

features). Each gene represents a specific feature as bit string. Each bit represents a

value one or zero for each feature. If the value 1 is represented in the bit, it means that

the feature is selected, otherwise the feature is not selected. The first bit refers to the

first feature; the second bit refers to the second feature and so on. Each chromosome is

represented by a binary vector of dimension F (where F is the total number of features),

as shown in the following Figure 3.

Figure 3: Structure of chromosome

S_F1 S_F2 S_F3 S_F4 S_F5 S_F6 S_F7 S_F8

1 1 0 0 1 1 0 1

1, selected feature

0, unselected feature if bit =

Features 1,2,5,6 and 8 are selected

Genetic Algorithm Structure

1. Initial population

Each individual is made up of a sequence of bits (0 and 1). Let N be the size of a

chromosome population. The population of 50 chromosomes is randomly generated in

the beginning. We used random function to generate a random floating-point array [0,

1]. We then used the round function to convert it into an integer, with a bias of 0.4 for

0 (random < 0.5) and 0.6 for 1 (random>=0.5).

2. Fitness function

The fitness value reflects how good a chromosome is compared to the other

chromosomes in the population. The higher chromosome has a higher chance of

survival and reproduction that can represent the next generation. In this paper, fitness(x)

is equal to the average recall from 70 training documents generated by ROUGE (Lin,

2004).

3. Selection

In this state, the existing population is selected to be a new generation through a fitness

based process. In each cycle, we chose two parents that give the highest average recall.

4. Crossover and mutation

Crossover

The function of the crossover is to generate new or child chromosomes from two parent

chromosomes by combining the information extracted from the parents. In each

generation, we generate new chromosomes using two crossover operations: one point

crossover and two point crossover. The one point crossover is chosen using a random

function. Then we swap the bit strings between two parents from the beginning until

the random point. One the other hand, in the two point of crossover, two random points

in the bit string is swapped with the parents from the first random point until the second

random point. The example is illustrated in Figure 4.

Figure 4: Generated next generation using crossover

Mutation

A mutation operator involves a probability that an arbitrary bit in a chromosome

sequence will be changed from an original state. In this paper, we generated new or

child chromosomes from two parent chromosomes by changing some parts of the

chromosome using bit swapping. We also used single mutation with 1, 2, 3, and 4

digit(s) for the bit swapping using random function. In Figure 5, an example for 2 digits

swapping, the two numbers were generated by random function. The random variable

tells us whether a particular bit will be modified. In Figure 5, the two random numbers

for the selected bits are 2 and 8, which means that the 2nd and 8th bits are swapped from

0 to 1 or 1 to 0.

Example, Random crossover point = 5

S_F1 S_F2 S_F3 S_F4 S_F5 S_F6 S_F7 S_F8

Parent1 1 0 1 1 1 1 1 0

Parent2 1 0 1 1 0 0 0 1

Child1 1 0 1 1 0 1 1 0

Child2 1 0 1 1 1 0 0 1

Figure 5: Generated next generation using mutation

These processes will continue until the fitness value of individuals in the population

converges.

5. Fitness Function Algorithm

The document sentences are scored using (eq.10) and ranked in a descending order

according to their scores. A set of the highest scoring sentences are extracted as a

document summary based on the compression rate. In this study, we used a 20%

compression rate as summary length. Then, we used an average recall generated by

ROUGE-1 (Lin, 2004) as the fitness function, as shown in equation 11

Where n is the length of the n-gram and countmatch is the highest number of n-grams

shared between a systems generated summary and a set of reference summaries.

Example, Random mutation point 2 and 8

S_F1 S_F2 S_F3 S_F4 S_F5 S_F6 S_F7 S_F8

Parent1 1 0 1 1 1 1 1 0

Parent2 1 0 1 1 0 0 0 1

Child1 1 1 1 1 1 1 1 1

Child2 1 1 1 1 0 0 0 0

Calculating Sentence Score

The individual chromosome selected is used to calculate score for each sentence in the

documents. The scores of each sentence feature are presented as a vector. The vectors

of the sentence features are used as inputs. Finally the sentence score is obtained using

this following equation (eq. 12)

Where Score(S) is the score of the sentence S , S_Fk(S) is the score of the feature k and

Ck is binary value of feature k.

We generate 100 generations and keep the highest fitness value from each generation

and then compare all highest fitness values. The best fitness value shows the features

that are suitable for a data set. The weights of the document features are calculated as

an average of the vectors created in each run. The final features weights are calculated

over the vectors of the features weights for all documents in the data collection.

After 100 documents were trained and tested by GA, the average weight of each feature

was obtained. We used the average weights to adjust the feature score for 62 unseen

documents. The sentence score for new documents can be calculated using the

following equation (13).

Where Score(S) is the score of the sentence S, is the average weight of the feature k

generated by GA and S_Fk(S) is the score of the feature k.

2.3 EXTRACTION OF SENTENCES

In those two methods, each sentence of the document is represented by a sentence

score. All document sentences are then ranked in descending order according to their

scores. A set of the highest scoring sentences are extracted as document summary based

on the compression rate. Therefore, we extracted the appropriate number of sentences

according to 20% compression rate. It has been proven that the extraction of 20% of

sentences from the source document can be considered as informative as the full text of

a document (Morris et al, 1992). Finally, the summary sentences are arranged in the

original order.

3.0 RESULTS

After the 100 documents were trained, we used 62 unseen documents to evaluate the

results using the ROUGE, a set of metrics called Recall-Oriented Understudy for

Gisting Evaluation. The evaluation toolkit (Lin, 2004) that has become a standard for

automatic evaluation of summaries. It compares the summaries generated by the

program with the human-generated (gold standard) summaries. For comparison, it uses

n-gram statistics. Our evaluation was done using n-gram setting of ROUGE, which was

found to have the highest correlation with human judgments at a confidence level of

95%. It is claimed that ROUGE-1 consistently correlates highly with human

assessments and has a high recall and precision significance test with manual evaluation

results. We choose ROUGE-1 as the measurement for our experimental results. In

Table 1, we compare the average precision, recall and f-measure score between general

statistic method (GSM), GA method, Microsoft Word 2007 Summarizer, and Copernic

summarizer.

Table 1: The comparison of average precision, recall and f-measure score among

four summarizers

Summari

zer

Training and Testing Data sets (100

Documents) 62 Unseen Documents

Avg.P Avg.R Avg.F Avg.P Avg.R Avg.F

GSM 0.49583 0.44216 0.46259 0.46646 0.44318 0.45170

GA Method 0.49800 0.44649 0.46622 0.46471 0.44673 0.45359

MS-Word 0.48189 0.39138 0.42279 0.46967 0.42265 0.43903

Copernic 0.51253 0.40984 0.44647 0.47131 0.42168 0.43975

We compare the results for both 100 documents of training-testing data sets and 62

unseen documents. In the 100 documents, the GSM reaches the average precision of

0.49583, recall of 0.44216 and f-measure of 0.46259. The GA method summarizer

achieves the average precision of 0.49800, recall of 0.44649 and f-measure of 0.46622.

While Microsoft Word 2007 summarizer reaches the average precision 0.48189, recall

of 0.39138 and f-measure of 0.42279 and the Copernic summarizer reaches an average

precision of 0.51253, recall of 0.40984 and f-measure of 0.44647.

We also compare the results of 62 unseen documents. The GSM gives the average

precision of 0.46646, recall of 0.44318 and f-measure of 0.45170. The GA method

summarizer achieves the average precision of 0.46471, recall of 0.44673 and f-measure

of 0.45359. While Microsoft Word 2007 summarizer reaches the average precision

0.46967, recall of 0.42265 and f-measure of 0.43903 and the Copernic summarizer

reaches an average precision of 0.47131, recall of 0.42168 and f-measure of 0.43975.

The results of the experiment in Figure 6 and 7 confirm that genetic algorithm has a

significant improvement in terms of quality of text summary. It is claimed that the

results of ROUGE-1 of all summarizers consistently correlate highly with human

assessments and have a high precision, recall and f-measure significance test with

evaluation results.

Figure 6: Average precision recall and f-measure score of 100 training documents

among four summarizers

Figure 7: Average precision recall and f-measure score of 62 unseen documents

among four summarizers

4.0 DISCUSSION

In this paper, we have presented a method based on general statistic method (GSM) and

a genetic algorithm aided sentence extraction summarizer that can be as informative as

the full text of a document with good information coverage. A prototype has also been

constructed to evaluate this automatic text summarization scheme by using some news

articles collection provided by DUC2002 as input. Afterwards, we extracted important

features and perform summary document based on those score called GSM. Then, we

proposed genetic algorithm (GA) method to adjust each feature. The benefit of GA is to

find and optimize the corresponding weight of each feature. Furthermore, GA is used to

obtain an appropriate set of feature weights. A chromosome is represented as the

combination of all feature weights. In training phrase, we defined fitness as the average

recall obtained with the genome. The average feature weights obtained by GA cannot

guarantee that the feature weights are best for the test corpus.

5.0 CONCLUSIONS

In this paper, we propose text summarization based on genetic algorithm to improve the

quality of summary results based on the general statistic method. We extracted the

important features for each sentence of the document and represent them as a vector of

features consisting of the following elements: title feature, sentence length, term weight,

sentence position, sentence to sentence similarity, proper noun, thematic word and

numerical data. We use 162 documents from DUC2002 data set. We divide 162

documents into two groups. The first 100 documents were used for training and testing.

The other 62 documents were used to evaluate and compare the results as unseen

documents. We compare our summarizer with the Copernic summarizer and Microsoft

Word 2007 summarizers. The results show that the genetic algorithm gives significant

quality improvement for text summarization.

ACKNOWLEDGEMENTS

This project is sponsored partly by the Ministry of Science, Technology and Innovation

under E-Science grant 01-01-06-SF0502, Malaysia. We would like to thank Suan Dusit

Rajabhat University and Universiti Teknologi Malaysia for supporting us.

REFERENCES

[1] DUC. Document understanding conference 2002, 2002. http://www.nlpir.nist.

gov/ projects/duc

[2] Fattah M.A. and Ren F. 2008. Automatic Text Summarization, In proceedings of

World Academy of Science, Engineering and Technology Volume 27.192-195.

[3] Kiani A. and Akbarzadeh, M.R. 2006. Automatic Text Summarization Using:

Hybrid Fuzzy GA-GP. In Proceedings of 2006 IEEE International Conference

on Fuzzy Systems, Sheraton Vancouver Wall Center Hotel, Vancouver, BC,

Canada. 977-983.

[4] Ko Y., Seo J. 2008. An effective sentence-extraction technique using contextual

information and statistical approaches for text summarization. Pattern

Recognition Letters archive, vol. 29, Issue 9, pp. 1366-1371. DOI:

10.1016/j.patrec.2008.02.008

[5] Luhn H. P. 1958. The Automatic Creation of Literature Abstracts. IBM Journal of

Research and Development, vol. 2. 159-165.

[6] Kupiec J., Pedersen, J. and Chen, F. 1995. A Trainable Document Summarizer. In

Proceedings of the Eighteenth Annual International ACM Conference on

Research and Development in Information Retrieval (SIGIR), Seattle, WA, 68-

73.

[7] Lin C.Y. 1999. Training a selection function for extraction. In Proceedings of the

eighth international conference on Information and knowledge management,

Kansas City, Missouri, United States. 55-62.

[8] Lin C.Y. 2004. ROUGE: A Package for Automatic Evaluation of Summaries. In

Proceedings of Workshop on Text Summarization of ACL, .Spain.

[9] Mani I. and Mark T. Maybury, (editors). 1999. Advances in automatic text

summarization MIT Press.

[10] Morris G., Kasper G.M., and Adam D.A. 1992. The effect and limitation of

automated text condensing on reading comprehension performance. Information

System Research, 3(1), 17-35.

[11] Porter M. F., 1980. An algorithm for suffix stripping. Morgan Kaufmann

Multimedia Information And Systems Series., Morgan Kaufmann Publishers Inc.

313-316. ISBN: 1-55860-454-5

[12] Ramiz M. Aliruliyev, 2009. A new sentence similarity measure and sentence

based extractive technique for automatic text summarization. Expert Systems

with Applications: An International Journal archive

36(4). 7764-7772. DOI: 10.1016/j.eswa.2008.11.022

[13] Salton G. 1989. Automatic Text Processing: The Transformation, Analysis, and

Retrieval of Information by Computer. Addison-Wesley Publishing Company.

[14] Salton G. and Buckley C. 1997. Term-weighting approaches in automatic text

retrieval. Information Processing and Management 24, 1988. 513-523. Reprinted

in: Sparck-Jones, K.; Willet, P. (eds.) Readings in I.Retrieval. Morgan

Kaufmann. 323-328.

[15] Srinivas M. and Patnaik L. M., 1994. Genetic algorithms: A Survey. Computer

27(6) (Jun. 1994), 17-26. DOI= http://dx.doi.org/10.1109/2.294849

[16] Suanmali, L., N. Salim and M.S. Binwahlan, 2009. Fuzzy Logic Based Method

for Improving Text Summarization. International Journal of Computer Science

and Information Security (IJCSIS), vol. 2(1), 65-70.

http://arxiv.org/abs/0906.4690

[17] Vahed, Q., Leila Sharif, H., and Ramin, H.,2008. Summarising text with a

genetic algorithm-based sentence extraction. International Journal of Knowledge

Management Studies, vol. 2(4), 426-444.

[18] Yeh, J.Y., H.R. Ke, W.P. Yang and I.H. Meng, 2005. Text summarization using a

trainable summarizer and latent semantic analysis. In: Special issue of

Information Processing and Management on An Asian digital libraries

perspective, 41(1), 75–95.

[19] Yulia, L., G. Alexander, and René Arnulfo García-Hernández, 2008. Terms

Derived from Frequent Sequences for Extractive Text Summarization. In: A.

Gelbukh (Ed.): CICLing 2008, LNCS vol. 4919, Springer, Heidelberg, 593-604.

DOI: 10.1007/978-3-540-78135-6

Mrs. Ladda Suanmali is a Ph.D. candidate in computer science in the Faculty of

Computer Science and Information Systems at Universiti Teknologi Malaysia. She

received her B.Sc. degree in computer science from Suan Dusit Rajabhat University,

Thailand in 1998, and her M.Sc. degree in Information Technology from King

Mongkut’s University of Technology Thonburi, Thailand in 2003. Since 2003, she has

been working as a lecturer at Suan Dusit Rajabhat University. Her current research

interests include text summarization, data mining, and soft computing.

Dr. Naomie Salim is an Assoc.Prof. presently working as a Deputy Dean of Postgraduate

Studies in the Faculty of Computer Science and Information System in Universiti

Teknologi Malaysia. She received her degree in Computer Science from Universiti

Teknologi Malaysia in 1989. She received her Master degree from University of Illinois

and Ph.D Degree from University of Sheffield in 1992 and 2002 respectively. Her current

research interest includes Information Retrieval, Distributed Database and

Chemoinformatic.

Dr. Mohammed Salem Binwahlan received his B.Sc. degree in Computer Science

from Hadhramout University of Science and Technology, Yemen in 2000. He received

his Master degree and Ph.D Degree from Universiti Teknologi Malaysia in 2006 and

2011 respectively. His current research interest includes Information Retrieval, Text

Summarization and Plagiarism Detection.